Physics Letters A 374 (2010) 4488–4495
Contents lists available at ScienceDirect
Physics Letters A www.elsevier.com/locate/pla
The dynamical origin of the universality classes of spatiotemporal intermittency Zahera Jabeen 1 , Neelima Gupte ∗ Department of Physics, Indian Institute of Technology-Madras, Chennai 600036, India
a r t i c l e
i n f o
Article history: Received 6 January 2010 Received in revised form 17 August 2010 Accepted 2 September 2010 Available online 6 September 2010 Communicated by A.R. Bishop
a b s t r a c t Two universality classes of spatiotemporal intermittency are seen in the spreading and non-spreading regimes of the sine circle map lattice, spatiotemporal intermittency of the directed percolation class, and spatial intermittency, not of the DP class, where the temporal behavior is regular. The transition between the two classes maps to a probabilistic to deterministic transition of the equivalent cellular automaton of the model, and is seen to have its dynamic origin in an attractor-widening crisis. © 2010 Elsevier B.V. All rights reserved.
Keywords: Spatiotemporal intermittency Spatial intermittency Directed percolation Cellular automaton Crisis
The identification of the universality class of spatiotemporal intermittency [1] in spatially extended systems has been a longstanding problem in the literature. Early conjectures argued that the transition to spatiotemporal intermittency is a second order phase transition, and the transition falls in the same universality class as directed percolation [2]. This conjecture has become the central issue in a long-standing debate [3–8], which is still not completely resolved. Studies of the coupled sine circle map lattice have thrown up a number of intriguing observations of relevance to this problem [9–11]. The phase diagram of the system contains spreading regimes where burst sites can infect their neighbors. This regime is separated by an infection line from a non-spreading regime where the bursts are non-infecting. Synchronized solutions can also be seen in these regions. Spatiotemporal intermittency (STI) with infecting bursts, and critical exponents which fall in the same universality class as directed percolation (DP) can be seen along the boundaries of synchronized solutions in the spreading regime. Conversely, spatial intermittency (SI) which has noninfecting bursts, and does not belong to the directed percolation class, can be seen along the boundaries of the synchronized solutions in the non-spreading regime.
Further insights into the spreading to non-spreading transition are obtained by mapping the coupled map lattice (CML) onto a cellular automaton [12]. The spreading to non-spreading transition seen across the infection line maps onto a transition from a probabilistic cellular automaton to a deterministic cellular automaton. Thus the change from spreading to non-spreading behavior seen in the CML is reflected in this transition. The probabilistic cellular automaton seen in the spreading regime exhibits scaling behavior and exponents consistent with the directed percolation universality class. The mean-field analysis of the cellular automaton mapping gives further insights into the probabilistic to deterministic cellular automaton transition, and also gives probability bounds for the directed percolation-like behavior. The origins of the spreading to non-spreading transition in the coupled map lattice, as well as the corresponding transition from probabilistic to deterministic behavior seen in the cellular automaton, lie in a dynamical phenomenon. The coupled map lattice undergoes an attractor-widening crisis at the infection line. The existence of this crisis can be inferred from the bifurcation diagram of the system and the distribution of finite time Lyapunov exponents. Thus the statistical characterizers of the system show signatures of a dynamical phenomenon. This result could have implications in a wider context. 1. The model and the universality classes
* 1
Corresponding author. Tel.: +91 44 2257 4861; fax: +91 44 2257 4852. E-mail address:
[email protected] (N. Gupte). Present address: Institute of Mathematical Sciences, Chennai, India.
0375-9601/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.physleta.2010.09.007
The coupled sine circle map lattice studied here is known to model the mode-locking behavior [13] seen in coupled oscillators,
Z. Jabeen, N. Gupte / Physics Letters A 374 (2010) 4488–4495
4489
Fig. 1. The figure shows the phase diagram of the coupled sine circle map lattice obtained with random initial conditions. The spreading and non-spreading regimes, separated by the infection line have been marked. Spatiotemporal intermittency of the directed percolation class is indicated by diamonds (3), whereas spatial intermittency has been marked with asterisks (∗) and triangles (). The ‘+’ signs locations where spatially intermittent solutions which are fixed points in time are seen.
Josephson junction arrays, etc. The model is defined by the evolution equation
xti +1 = (1 − ε ) f xti +
ε
f xti−1 + f xti+1 2
(mod 1)
(1)
where i and t are the discrete site and time indices respectively and ε is the strength of the coupling between the site i and its two nearest neighbors. The local on-site map, f (x) is the sine circle map defined as f (x) = x + Ω − 2Kπ sin(2π x), where K is the strength of the nonlinearity and Ω is the winding number of the single sine circle map in the absence of the nonlinearity. We study the system with periodic boundary conditions in the parameter regime 0 < Ω < 21π (where the single circle map has temporal period 1 solutions), 0 < ε < 1 and K = 1.0. The phase diagram of this model evolved with random initial conditions is shown in Fig. 1. A large part of the phase diagram is occupied by spatiotemporally fixed point solutions (indicated by dots), in which all the sites relax to the fixed point x = 21π sin−1 ( 2πKΩ ). Additionally, two distinct regimes, separated by a line, the infection line, can be seen in the phase diagram. A spreading regime where the burst states can infect their neighboring laminar states and spread through the lattice can be seen above the infection line (Fig. 2(a)). (Here, the sites which relax to the fixed point solution x , are identified as the laminar sites.) This is in contrast to the non-spreading regime seen below the infection line, where the random initial conditions die down to bursts which are localized, and do not infect their neighboring laminar states (Fig. 2(b)). In both these regions, spatiotemporal intermittency with coexisting laminar states exhibiting regular temporal behavior, and turbulent states with irregular temporal behavior, can be identified near the bifurcation boundary of spatiotemporally fixed point solutions. The spatiotemporal intermittency seen in the spreading regime has been shown to belong to the directed percolation class. Some of the points that show this have been marked with diamonds (3) in the phase diagram (Fig. 1). An entire set of static and dynamic scaling exponents was obtained in this parameter regime, which showed good agreement with the universal DP exponents [9,10]. These are summarized in Table 2 in Appendix A together with the definitions of the exponents. These exponents also satisfy the hyperscaling relations for the static exponents (2β/ν = d − 2 + η ), and the spreading exponents (4δ + 2η = dz s ) for d = 1. Hence, this class of spatiotemporal intermittency seen at the bifurcation boundary in the spreading regime belongs convinc-
Fig. 2. (Color online.) The figure shows the space–time plot of (a) spatiotemporal intermittency seen at Ω = 0.06, ε = 0.7928 (one of the points marked with diamonds in Fig. 1), and (b) spatial intermittency seen at Ω = 0.047, ε = 0.336 (marked with asterisk in Fig. 1). The index i represents the lattice sites and the index t represents time.
ingly to the directed percolation class. These scaling exponents are however, seen only near the bifurcation boundary. The distribution of laminar lengths for the spreading solutions off the bifurcation boundary, show an exponential fall off. Thus STI of the DP class is a special case of the spatiotemporal behavior seen in the spreading regime. A different class of intermittency, namely spatial intermittency, is seen near the bifurcation boundary in the non-spreading regime [9–11]. In this class of intermittency, the laminar sites are the synchronized fixed point x∗ and the temporal behavior of the burst states is either quasi-periodic or periodic (Fig. 2(b)). Some of these points have been marked with triangles (, quasi-periodic) and asterisks (∗, periodic) in the phase diagram (Fig. 1). In this class of intermittency, the distribution of laminar lengths, obtained by averaging over different random initial conditions, scales as a powerlaw of the form P (l) ∼ l−ζ , with ζ as the associated scaling expo-
4490
Z. Jabeen, N. Gupte / Physics Letters A 374 (2010) 4488–4495
nent [11]. The exponent ζ obtained in this case was found to be ∼ 1.1 (see Table 2). A similar scaling exponent has been seen in the case of the spatial intermittency observed in inhomogeneous logistic map lattice [14]. The laminar length distributions seen for other non-spreading solutions at points off the bifurcation boundary show an exponential fall off. Thus, the spatially intermittent solutions are special cases in the non-spreading regime. Thus the sine circle coupled map lattice shows a transition from a spreading regime to a non-spreading regime at the infection line, which in turn is responsible for the presence of both directed percolation-like spatiotemporal intermittency and spatial intermittency in the same system. In order to gain further insights into this transition, we map the coupled map lattice to a stochastic model, a probabilistic cellular automaton of the Domany–Kinzel type [3,15].
Table 1 The table shows the probabilities pk ’s obtained in the spreading (S), nonspreading (NS) regimes and at directed percolation (DP) and spatial intermittency (SI) points.
Ω
ε
p0
p1
p2
p3
p4
p5
S (DP)
0.060 0.073
0.7928 0.4664
0 .0 0 .0
0.21 0.15
0 .0 0 .0
0.90 0.94
0.57 0.44
0.97 0.99
S
0.070
0.264 0.248
0 .0 0 .0
0.14 0.05
0 .0 0 .0
0.98 0.99
0.39 0.16
0.99 0.99
NS
0.070
0.232 0.228
0 .0 0 .0
0.00 0.00
0 .0 0 .0
1.00 1.00
0.00 0.00
1.00 1.00
NS (SI)
0.031 0.044
0.420 0.373
0 .0 0 .0
0.00 0.00
0 .0 0 .0
1.00 1.00
0.00 0.00
1.00 1.00
2. Mapping to an equivalent cellular automaton Signatures of the spreading to non-spreading transition at the infection line are seen in a mapping of the coupled map lattice to a stochastic model namely, a probabilistic cellular automaton of the Domany–Kinzel type [3,15]. The equivalent cellular automaton, set up to mimic the dynamics of the laminar and burst states in the coupled map lattice defined in Eq. (1), is defined on a onedimensional lattice of size N [12]. The state variable v ti at site i and at time t is assigned the value v ti = 0 if the site is in the laminar state, and v ti = 1 if the site is in the burst state. As in the coupled map lattice equation (1), the probability of the site i at time t + 1 being in the burst state depends on the state of the sites i − 1, i and i + 1 at time t. We therefore define the cellular automaton dynamics in this system by the conditional probability P ( v ti +1 | v ti−1 , v ti , v ti+1 ). There are 23 possible configurations and the symmetry between the sites i − 1 and i + 1 in Eq. (1) gives the effective probabilities pk ’s as p 0 = P (1|000), p 1 = P (1|001) = P (1|100), p 2 = P (1|010), p 3 = P (1|011) = P (1|110), p 4 = P (1|101) and p 5 = P (1|111) [16]. These probabilities then define the update rules of the cellular automaton. We estimate these probabilities pk , for a given set of parameter values, from the numerical evolution of the coupled map lattice with random initial conditions. The probabilities pk ’s are calculated by finding the fraction of sites i which exist in the burst state v ti +1 = 1 at time t + 1, given that the site i and its nearest neighbors i − 1 and i + 1 existed in state k at time t. That is, the probability pk is determined using pk = N 1k
N 1k
N 0k + N 1k
, where N 0k
and are the number of sites, which at time t were the central sites of the configuration k, and at time t + 1 exist in the laminar states (v ti +1 = 0) and the burst states (v ti +1 = 1) respectively. These probabilities, were extracted from a coupled map lattice of size N = 103 averaged over 14,000 time steps discarding a transient of 1000 time steps, and averaged over 200 initial conditions. The probabilities pk obtained in the spreading and non-spreading regimes in the phase diagram are listed in Table 1. It is clear that the condition for an absorbing state is realized in the probability p 0 = P (1|000) which is seen to be zero in both regimes. The probabilities p 1 = P (1|001) = P (1|100) and p 4 = P (1|101) essentially define infection probabilities, which estimate the probability of a laminar site being infected by its burst neighbor or neighbors, to change to a burst site. We can see from Table 1 that these probabilities show drastically different behavior in the spreading and non-spreading regimes. In the case of the spreading regime, the probabilities obtained are seen to lie in the open interval (0, 1). Therefore, the dynamics in the spreading regime is described by a probabilistic cellular automaton (PCA) wherein the cellular automaton rules are probabilistic in nature. In contrast, in the case of the non-spreading
Fig. 3. (Color online.) The figure shows the probabilities associated with the update rules p 1 , p 3 , p 4 , p 5 plotted as a function of the coupling strength, ε at Ω = 0.065. In the non-spreading regime (ε < εic ), the probabilities are either 0 or 1, whereas they lie in the interval (0, 1) in the spreading regime (ε εic ).
regime, in addition to p 0 and p 2 which are zero for the STI of the DP type [17], the infection probabilities p 1 and p 4 are also seen to be equal to zero. Hence, the infection probabilities characterize the non-infective nature of the bursts. Moreover, the probabilities p 3 = P (1|011) = P (1|110) and p 5 = P (1|111) take the value 1, which indicates the robustness of the burst states. We note that the probabilities obtained here only take the values 0 or 1. Hence, we obtain a deterministic cellular automaton (DCA) in the nonspreading regime, wherein given a state k at time t, the state of the site i at time t + 1 is decisively known with probability zero or one. This is further illustrated in Fig. 3, where the probabilities have been plotted as a function of the coupling strength ε at Ω = 0.065. The probabilities show a distinct change to values other than zero or one, at the point where spreading regime starts viz. ε > εic (see inset in Fig. 1). Thus a transition from a probabilis-
Z. Jabeen, N. Gupte / Physics Letters A 374 (2010) 4488–4495
4491
Fig. 4. (Color online.) The figure shows the log–log (base 10) plot of physical quantities calculated for the probabilistic cellular automaton associated with the DP point Ω = 0.06, ε = 0.7928. The figure shows the (a) escape time τ vs length of the lattice L, (b) order parameter m vs time t, and the spreading properties (c) the fraction of burst sites N (t ) vs t, and (d) the survival probability P (t ) plotted as a function of time t. The fits to the plots and the corresponding scaling exponents (error bars in brackets) obtained are also shown.
tic to a deterministic cellular automaton is seen at the infection line. It is therefore clear that the spatiotemporal intermittency of the directed percolation class seen in the spreading regime can be modeled by a PCA, whereas spatial intermittency observed in the non-spreading regime can be modeled by a DCA. In the following section, we show that the probabilistic cellular automata obtained at the STI of the DP points indeed mimics the dynamics of the coupled map lattice and shows scaling exponents which match with the universal directed percolation exponents. 2.1. Probabilistic cellular automaton and directed percolation As mentioned previously, the dynamics seen in the case of spatiotemporal intermittency in the spreading regime resembles that of directed percolation. The stochastic nature of the burst states is further emphasized in the probabilistic cellular automaton obtained in this regime. In this section, we show that the probabilistic cellular automata obtained at points in the phase diagram which show DP-like spatiotemporal intermittency, indeed show scaling behavior which match with the directed percolation class. Fig. 4 shows some of the physical quantities associated with the static and dynamical properties of the system. The scaling exponents obtained for these quantities are also shown in the figure. The probabilities associated with the PCA obtained at the parameter value Ω = 0.06, ε = 0.7928, were found to be p 0 = 0.0, p 1 = 0.213116, p 2 = 0.000081, p 3 = 0.902515, p 4 = 0.569464 and p 5 = 0.967595. When the cellular automaton is prepared with random initial conditions, such that the sites take either 0 or 1 values with equal probabilities at time t = 0, the burst state (1 state) is seen to die down with time. The relaxation time τ , which determines the time taken by the lattice to relax to a laminar state (0 state), scales as a function of the size of the lattice L as τ ∼ L z (Fig. 4(a)). The order parameter m, defined as the fraction of burst sites in the lattice, decreases with time as m ∼ t −β/ν z (Fig. 4(b)). In contrast, when a pair of burst states are introduced in a lattice prepared in the laminar state at time t = 0, the burst states are
seen to grow with time. The dynamical quantities associated with the growth of the burst states viz. the fraction of burst sites N (t ), and the survival probability P (t ), which is defined as the fraction of initial conditions which yield a non-zero number of burst states at time t, are shown in Fig. 4(c) and (d). The respective scaling exponents obtained show good agreement with the directed percolation class where the values of the directed percolation exponents are listed in Table 2. Thus the cellular automaton model obtained in the region of spatiotemporal intermittency seen in the spreading regime exhibits scaling behavior similar to the directed percolation class, as expected. We note that in the case of the non-spreading regime, the deterministic cellular automaton obtained acts like an identity mapping. Therefore, the scaling behavior in the laminar length distribution depends strongly on the initial condition namely, the fraction of burst states in the lattice. To mimic the dynamics exhibited by the coupled map lattice, we calculate the fraction of burst sites in a coupled map lattice of size 20,000 at Ω = 0.04, ε = 0.402, after discarding 60,000 transients and averaging over 50 random initial conditions. The fraction of burst sites came out to be p = 0.0684. Fig. 5 shows the scaling behavior of the distribution of laminar lengths for the deterministic cellular automaton, obtained for this initial condition, in which bursts were introduced with probability p. The exponent seen in the cellular automaton turns out to have the value ζ ∼ 1.1 as in the coupled map lattice. A mean-field study of the cellular automaton gives further insights into the transition from probabilistic to deterministic cellular automaton, as well as into the initial condition dependence of the cellular automata. We discuss this in the subsequent section. 2.2. Mean-field analysis of the cellular automaton We study a mean field approximation of the probabilistic cellular automaton which gives a better understanding of the probabilistic to deterministic cellular automaton transition as well as gives probability bounds for the directed percolation-like behavior [18]. Let mt and mt +1 be the density of burst states in the lattice
4492
Z. Jabeen, N. Gupte / Physics Letters A 374 (2010) 4488–4495
Fig. 6. (Color online.) The figure shows the stability regions of the three fixed points of the mean-field equation (3), m1 , m2 , and m3 , in the { p 1 , p 3 , p 4 } phase space. The regions in which mi are stable are indicated by the arrows. The probabilities of the DP points (marked with diamonds), as well as for points in the spreading regime at Ω = 0.065 (dotted line) have been shown. The triangle corresponds to the probabilities in the non-spreading regime. Fig. 5. The figure shows the laminar length distribution obtained for the deterministic cellular automaton obtained at the spatial intermittency point Ω = 0.04, ε = 0.402. The exponent obtained is ζ = 1.106 ± 0.005.
at the tth and (t + 1)th time step. The mean-field equation for the probabilistic cellular automaton, based on the update rules defined earlier in this section, can be written as
mt +1 = (2p 1 + p 2 )mt (1 − mt )2
+ (2p 3 + p 4 )mt2 (1 − mt ) + p 5mt3
(2)
By approximating p 5 = 1 and p 2 = 0, the mean field equation reduces to
mt +1 = 2p 1mt (1 − mt )2 + (2p 3 + p 4 )mt2 (1 − mt ) + mt3
(3)
The three fixed points of this equation for a set of parameters 1 −1 { p 1 , p 3 , p 4 } are m1 = 0, m2 = 1+2p2p , and m3 = 1. A linear 1 −2p 3 − p 4 stability analysis of the mean-field equation indicates that the fixed point m1 = 0, which corresponds to the absorbing state, is stable
when p 1 < 12 . The fixed point m2 is stable when 2p 3 + p 4 2 and 2p 1 1. The fixed point m3 = 1, which corresponds to the completely turbulent state, is stable when 2p 3 + p 4 2. The stability regions of these three fixed points in the { p 1 , p 3 , p 4 } space have been indicated in Fig. 6. The 2p 3 + p 4 = 2 plane, above which m3 = 1 solutions are found to be stable, is shown in the figure. We see that both m1 = 0 and m3 = 1 solutions are stable, when
p 1 12 and 2p 3 + p 4 2. This is the co-existence region. The system settles down to either the completely laminar state (m1 = 0) or the completely turbulent state (m3 = 1) in the co-existence region, depending on the initial condition mt at t = 0. The solutions settle down to the m1 = 0 solution if initial density, m0 ∈ [0, m2 ) whereas, they tend towards m3 = 1 if m0 ∈ (m2 , 1]. For instance, we iterate the map equation (3) by choosing p 4 = 0.9 and study the {m, p 1 , p 3 } phase diagram for two initial conditions m0 = 0.1 and m0 = 0.9. The {m, p 1 , p 3 } phase diagram obtained is shown in Fig. 7(a) and (b) respectively. We see that the solutions settled down to m = 0 in the co-existence region in the first case, whereas they settle down to m = 1 in the latter case. Hence, the solutions in the co-existence region are initial condition dependent. Interestingly, we find that the probabilities associated with the spreading regime of the coupled map lattice, including those obtained for the directed percolation points in the spreading regime, lie in the co-existence region (Fig. 6). Due to this, the probabilistic cellular automaton has a very strong initial condition dependence in the spreading region. The correct choice of initial conditions leads to the laminar absorbing state. Other choices can end up easily in the m = 1 state. However, the points at which directed
percolation is seen are exceptions to this. The probabilistic cellular automata obtained at these points show scaling behavior consistent with the directed percolation phenomena irrespective of the choice of initial conditions, as was discussed in the previous section. The probabilities associated with the deterministic cellular automaton seen in the non-spreading regime (i.e. p 1 = 0, p 3 = 1, p 4 = 0) lie at the vertex belonging to the co-existence region in this cube. This point has been marked with a triangle in Fig. 6. As we approach the spatial intermittency points along the bifurcation boundary, the cellular automaton probabilities obtained at the directed percolation points tend towards this vertex. Similarly, the probabilities obtained in the spreading regime tend towards this vertex, as we cross the infection line and enter the non-spreading regime. Therefore, the deterministic cellular automaton is a limiting case of the probabilistic cellular automaton seen in the spreading regime. This confirms that the presence of both spreading and nonspreading regimes separated by the infection line in the phase diagram, which accounts for the presence of both directed percolation-like spatiotemporal intermittency and spatial intermittency in the phase diagram, is manifested in the form of a transition from probabilistic to deterministic cellular automaton at the infection line. However, the origins of the spreading nature of the burst states above the infection line are not clear. In the next section, we identify the dynamical origins of the spreading states. 3. Crisis at the infection line It is clear from the preceding discussion that the behavior of the burst states changes drastically at the infection line leading to the spreading to non-spreading transition and the existence of distinct classes of spatiotemporal intermittency in the phase diagram. We now have indications that the origin of the spreading states lies in an attractor-widening crisis at the infection line, as we will see in this section. A crisis is said to have occurred in a dynamical system when a sudden change in the attractor takes place when a system parameter is changed [19]. We observe such a phenomenon in our coupled map lattice, when we change the strength of the coupling between sites, ε , for a given Ω in the vicinity of the infection line. This has been illustrated in Fig. 8. Here, the variable x associated with a typical site has been plotted over 500 time steps, as a function of the coupling strength ε at Ω = 0.065. The range of ε values on the vertical axis of Fig. 1 cuts across the infection line at εic = 0.252 for Ω = 0.065. The bifurcation diagram clearly shows that an attractor-widening crisis [20] ap-
Z. Jabeen, N. Gupte / Physics Letters A 374 (2010) 4488–4495
4493
Fig. 7. The figure shows the {m, p 1 , p 3 } phase diagram of the mean field map at p 4 = 0.9 with initial m at time t = 0 chosen as (a) m0 = 0.1 and (b) m0 = 0.9. In the co-existence region (2p 3 1.1 and p 1 0.5), the density of burst sites m, settles down to either m1 or m3 depending on the initial density m0 chosen.
Fig. 8. The figure shows the bifurcation diagram of the coupled map lattice in which the variable x associated with a typical site has been plotted over 500 time steps, as a function of the coupling strength ε in the neighborhood of the infection line, at Ω = 0.065. The lattice size is L = 200. A transient of 510,000 iterates has been discarded.
pears at this point. Similar behavior is seen for other sites. The spreading regime seen in the phase diagram emerges exactly at the point at which the attractor widens, with the non-spreading regime corresponding to the pre-widening regime. This widening also identifies the point at which the equivalent cellular automaton undergoes a probabilistic to deterministic transition. In the precrisis region, each site follows either a periodic or quasi-periodic trajectory and is not infected by the behavior of its neighbors. Thus, its CA analogue is deterministic as listed in Table 1. In the post-crisis regime,each site is able to access the full x range, as well as infect its neighbors, and the bursting and spreading behavior characteristic of the spreading regime is seen. This is reflected in the equivalent cellular automaton by a transition to probabilistic behavior (Table 1). It is to be noted that the volume of the attractor in phase space will be much larger post-crisis, as compared to the pre-crisis volume. Hence, we see that the spreading nature of the bursts above the infection line can be attributed to the attractor-widening crisis at the infection line. We also see indications of unstable dimension variability in the vicinity of the infection line. A chaotic attractor is said to have unstable dimension variability, if it contains embedded periodic orbits with different number of stable and unstable directions
[21,22]. A trajectory visiting the neighborhood of these periodic orbits, experiences a fluctuating number of unstable directions, as time progresses. The signature of this phenomenon can be seen in the finite time Lyapunov exponents (FTLE) of the system which fluctuate about zero, as the dynamics evolves. We find that the finite time Lyapunov exponents obtained for our system exhibit fluctuations about zero in the vicinity of the infection line. This can be seen in Fig. 9(a), where the fraction of positive time-20 Lyapunov exponents obtained at Ω = 0.065, ε = 0.254, is plotted as a function of time. The fraction of positive exponents is seen to vary with time, indicating that a fraction of the Lyapunov exponents fluctuate about zero, as the dynamics evolves. Hence, in the vicinity of the infection line, the trajectory of the system visits local regions which contain unstable periodic orbits with different numbers of unstable directions. This is further illustrated in Fig. 9(b), which shows the distribution of the largest time-100 Lyapunov exponent, obtained at Ω = 0.065 for various values of ε . The distribution has been calculated by collecting data for a lattice of size N = 200 over 10,000 time steps and by averaging over 200 initial conditions. We find that the distribution of the largest time-100 Lyapunov exponent is constrained to the negative values for parameters lying in the non-spreading regime (e.g. ε = 0.245 in Fig. 9(b)). For parameters close to the infection line (viz. ε = 0.252), the distribution shifts towards the positive side, until it lies entirely in the positive side of the axis for parameters above the infection line (e.g. ε = 0.260). This behavior is typically seen in the neighborhood of a crisis [23] and supports the behavior observed in the bifurcation diagram. 4. Discussion To conclude, the origins of the two universality classes of spatiotemporal intermittency seen in the phase diagram of the sine circle map lattice (i.e. the directed percolation class of STI and the non-directed percolation class of spatial intermittency), lie in a dynamical phenomenon, viz. an attractor-widening crisis which occurs at the parameter values which lie on the infection line which separates the spreading and non-spreading regimes in the phase diagram. Thus, the attractor-widening crisis is intimately connected with the spreading to non-spreading transition and the existence or suppression of the spreading or infectious modes. This connection is also seen in the coarse grained cellular automaton description of the system which undergoes a transition from probabilistic behavior to deterministic behavior at the infection line, with a corresponding suppression of infectious modes.
4494
Z. Jabeen, N. Gupte / Physics Letters A 374 (2010) 4488–4495
Thus, in this system, there is a direct connection between a dynamical phenomenon viz. a crisis in an extended system and the statistical properties of the extended system viz. the exponents and universality classes. Similar directed percolation to nondirected percolation transitions have been seen in other coupled map lattices [24,25], as well as in pair contact processes [26], solid on solid models [27], branching-annihilating random walks [28, 29] and models of non-equilibrium wetting [30,31]. Some of these models are inspired by studies of the synchronization transition in CML-s [32–36] where transitions like the KPZ transition and the anomalous directed percolation transition are also seen. Thus transitions similar to those seen in our model are seen in CML models, cellular automata and in non-equilibrium statistics.
Our results clearly indicate that it is useful to look at such transitions from more than one point of view, an evolving large dimensional dynamical system such as a CML, a coarse grained equivalent cellular automaton, and a statistical description in terms of exponents, as each description sheds light on a different aspect of the transition, and hence on the behavior of the system. It would be interesting to see if such equivalent descriptions can yield insights into the behavior of other systems, e.g. systems which show the synchronization transition, or non-equilibrium systems, and thus contribute to the on-going debate on the identification of the universality classes of spatiotemporal systems. Acknowledgement N.G. thanks DST, India for partial support under the project SR/S2/HEP/10/2003. Appendix A. Directed percolation exponents The directed percolation transition is characterized by a set of static and dynamic critical exponents associated with various quantities of physical interest [37] (Table 2). The static exponents are defined as follows: (i) The escape time τ (Ω, ε , L ), defined as the time taken for the system starting from random initial conditions to relax to a completely laminar state, varies with the system size L as τ ∼ L z at criticality (ε = εc ), with z as the associated exponent. (ii) The order parameter m(t ), defined as the fraction of turbulent sites in the lattice at time t, scales as m ∼ (ε − εc )β for ε → εc+ . At t τ , m(t ) scales with t as m(εc , t ) ∼ t −β/ν z , where ν is exponent associated with the spatial correlation length. (iii) The correlation function in space is defined as
1 L
C j (t ) =
L
i =1
2
xti xti+ j − xti
At εc , C j (t ) scales as C j (t ) ∼ j 1−η . (iv) The distribution of laminar lengths, P (l) shows a power-law behavior of the form P (l) ∼ l−ζ . ζ is the associated exponent. To extract the dynamical exponents, two turbulent seeds are placed in an absorbing lattice and the spreading of the turbulence in the lattice is studied. The quantities associated with critical exponents at εc are
Fig. 9. (Color online.) The figure shows (a) the fraction of positive finite time Lyapunov exponents (time = 20) as a function of time t obtained at Ω = 0.065, ε = 0.254, and (b) the distribution of the largest finite time Lyapunov exponent (time = 100) plotted for Ω = 0.065 and ε = 0.245, 0.252, 0.260.
Table 2 The static and dynamic (spreading) exponents obtained at two of the directed percolation (DP) points (3-s) in Fig. 1 are shown in the first two rows of the table [10,11]. The universal DP exponents are listed in the third row. The laminar length distribution exponent, ζ , calculated for spatial intermittency (SI) at two points marked by -s (quasi-periodic bursts) and ∗-s (periodic bursts) respectively in Fig. 1, are also listed. The error-bars are shown in the brackets. The data has been obtained for a lattice of size, N = 103 and has been averaged over 103 initial conditions. The laminar length distributions have been calculated for a lattice of size N = 104 and averaged over 50 initial conditions.
εc (Ω)
Ω STI-DP
Spreading exponents
z
β/ν z
β
η
ζ
η
δ
zs
0.17 (0.02) 0.16 (0.01) 0.16
0.293
1.68 (0.01) 1.66 (0.01) 1.67
0.315 (0.007) 0.303 (0.001) 0.313
0.16 (0.01) 0.16 (0.01) 0.16
1.26 (0.01) 1.27 (0.01) 1.26
1.10 (0.04) 1.13 (0.02)
–
–
–
–
–
–
0.060
0.7928
0.065
0.34949
1.59 (0.02) 1.59 (0.03) 1.58
0.28
1.51 (0.01) 1.50 (0.01) 1.51
0.04
0.402
–
–
–
–
0.047
0.336
–
–
–
–
DP SI
Static exponents
0.273
Z. Jabeen, N. Gupte / Physics Letters A 374 (2010) 4488–4495
(i) the number of active sites, N (t ) at time t, which scales as N (t ) ∼ t η , (ii) the survival probability, P (t ) defined as the fraction of initial conditions which show a non-zero number of active sites at time t. This scales as P (t ) ∼ t −δ , and (iii) the radius of gyration R 2 (t ), which is defined as the mean squared deviation of the position of active sites from the original sites of turbulent activity. This scales as R 2 (t ) ∼ t zs . References [1] F. Daviaud, M. Bonetti, M. Dubois, Phys. Rev. A 42 (1990) 3388; P.W. Colovas, C.D. Andereck, Phys. Rev. E 55 (1997) 2736; M. Das, B. Chakrabarti, C. Dasgupta, S. Ramaswamy, A.K. Sood, Phys. Rev. E 71 (2005) 021707; C. Pirat, A. Naso, J.L. Meunier, P. Massa, C. Mathis, Phys. Rev. Lett. 94 (2005) 134502. [2] Y. Pomeau, Physica D 23 (1986) 3. [3] H. Chaté, P. Manneville, Physica D 32 (1988) 409. [4] P. Grassberger, T. Schreiber, Physica D 50 (1991) 177. [5] T. Bohr, M. van Hecke, R. Mikkelsen, M. Ipsen, Phys. Rev. Lett. 86 (2001) 5482. [6] J. Rolf, T. Bohr, M.H. Jensen, Phys. Rev. E 57 (1998) R2503. [7] P. Rupp, R. Richter, I. Rehberg, Phys. Rev. E 67 (2003) 036209. [8] K.A. Takeuchi, M. Kuroda, H. Chaté, M. Sano, Phys. Rev. Lett. 99 (2007) 234503. [9] T.M. Janaki, S. Sinha, N. Gupte, Phys. Rev. E 67 (2003) 056218. [10] Z. Jabeen, N. Gupte, Phys. Rev. E 72 (2005) 016202. [11] Z. Jabeen, N. Gupte, Phys. Rev. E 74 (2006) 016210. [12] Z. Jabeen, N. Gupte, Physica A 384 (2007) 59. [13] G. Pradhan, N. Chatterjee, N. Gupte, Phys. Rev. E 65 (2002) 046227. [14] A. Sharma, N. Gupte, Phys. Rev. E 66 (2002) 036210.
4495
[15] E. Domany, W. Kinzel, Phys. Rev. Lett. 53 (1984) 311. [16] PCA-DCA regimes have also been seen in the Chaté Manneville CML. See H. Chaté, P. Manneville, J. Stat. Phys. 56 (1989) 357. [17] The probability p 2 = P (1|010) is also seen to be equal to zero in both the regimes, though the reasons are different. In the spreading regime, we find that the neighboring laminar states suppress the central burst state, and hence we obtain a zero probability for p 2 . In the case of the non-spreading regime, a single burst site with two laminar neighbors is never observed. [18] A similar study has been carried out in F. Bagnoli, et al., Phys. Rev. E 63 (2001) 046116, in which the probability rules were defined differently. [19] C. Grebogi, E. Ott, J.A. Yorke, Phys. Rev. Lett. 48 (1982) 1507. [20] C. Grebogi, E. Ott, F. Romeiras, J.A. Yorke, Phys. Rev. A 36 (1987) 5365. [21] E.J. Kostelich, I. Kan, C. Grebogi, E. Ott, J.A. Yorke, Physica D 109 (1997) 81. [22] Y.C. Lai, C. Grebogi, J. Kurths, Phys. Rev. E 59 (1999) 2907; Y.C. Lai, D. Lerner, K. Williams, C. Grebogi, Phys. Rev. E 60 (1999) 5445. [23] K.T. Alligood, E. Sander, J.A. Yorke, Phys. Rev. Lett. 96 (2006) 244103; G.T. Kubo, R.L. Viana, S.R. Lopez, C. Grebogi, Phys. Lett. A 372 (2008) 5569. [24] P. Marcq, H. Chaté, P. Manneville, Phys. Rev. Lett. 77 (1996) 4003. [25] P. Marcq, H. Chaté, P. Manneville, Phys. Rev. E 55 (1997) 2606. [26] H. Hinrichsen, Physica A 361 (2006) 457. [27] A. Lipowski, Michel Droz, Phys. Rev. E 68 (2003) 056119. [28] J. Kockelkoren, H. Chaté, Phys. Rev. Lett. 90 (2004) 125701. [29] G. Ódor, N. Menyhárd, Phys. Rev. E 73 (2006) 036130. [30] F. Ginelli, V. Ahlers, R. Livi, D. Mukamel, A. Pikovsky, A. Politi, A. Torcini, Phys. Rev. E 68 (2003) 065102(R). [31] G. Odor, Rev. Mod. Phys. 76 (2004) 663. [32] M. Droz, A. Lipowski, Phys. Rev. E 67 (2003) 056204. [33] A.S. Pikovsky, J. Kurths, Phys. Rev. E 49 (1994) 898. [34] V. Ahlers, A. Pikovsky, Phys. Rev. Lett. 88 (2002) 254101. [35] M. Cencini, C.J. Tessone, A. Torcini, Chaos 18 (2008) 037125. [36] M.A. Muñoz, R. Pastor-Satorras, Phys. Rev. Lett. 90 (2003) 204101. [37] D. Stauffer, A. Aharony, Introduction to Percolation Theory, Taylor and Francis, London, 1992.